Create images like thekumarmethod from your own video | TheFluxTrain

You have a clip where the movement is perfect—gesture, timing, camera—but the person on screen is wrong. Maybe you need a different face, updated on-screen text, or a fresh identity for the same performance.

Reshooting is expensive. A one-shot “change this video” prompt usually fails because video models rewrite motion when you ask for appearance edits.

The Kumar Method (this Flow Studio template) splits the problem into two jobs:

1. Lock the motion from your source clip.
2. Restyle one clear frame with GPT Image 2—swap face, tweak text, or adjust styling—while keeping pose, outfit, framing, and lighting.

This tutorial covers step 1 and the still-frame restyle. When you are happy with the frame, you can feed it into Motion Apply so the original movement drives the new look.

Note: Image and video generations use credits. Expect a few runs while you pick the best frame and restyle result—especially if hands, skin tone, or small text need another pass.

Try it: The Kumar Method in Explore

Full video (motion + tail montage): /tutorials/flow-studio-thekumarmethod-full-video-tutorial

What the finished still looks like

Sample GPT Image 2 — Restyle Frame outputs from the template sink—arrow through variants in Flow Studio and pick the one where face, hands, and text look right.

What you’ll need

• Source video — a clip with the motion you want to keep (upload or paste a link).
• Reference photo (optional but recommended) — a clear headshot or portrait for the new identity.
• Short edit note — plain language for what should change, for example “Change text to TheFluxTrain” or “Make the person look like this reference.”

Time: First run usually takes 5–15 minutes including uploads and one or two restyle attempts.

The workflow in one sentence

Upload video → grab the clearest frame → describe what to change → GPT Image 2 restyles the still while preserving pose and scene.

How the graph is wired (plain English)

The Instructions note on the canvas repeats these steps for quick reference while you work.

Step-by-step

1. Open the template

Go to The Kumar Method in Explore. The graph loads with demo media so you can explore before swapping in your own files.

2. Add your source video

Select 1. Source Video and replace the placeholder with your clip—upload from disk or paste a hosted URL.

Pick footage where the subject’s face is visible, not motion-blurred, and not blocked by props or hands.

3. Extract the best frame

Select 2. Extract First Frame and click Generate.

The node shows Start, Custom, and End frame previews. Choose the one where the person reads clearest—usually Custom after you set Custom time to a moment mid-gesture.

If the Custom frame is wrong, adjust Custom time slightly and generate again. Do not move on until one frame looks sharp.

4. Add a reference photo (optional)

Replace the Character reference image with your target identity—a well-lit portrait works best.

This image feeds Image 2 on the GPT Image 2 node so the swap has a concrete face to match.

5. Write your edit in plain language

Open Additional edits and type what should change. Keep it short and visual—appearance edits, not new actions.

Example — on-screen text:

Change text to TheFluxTrain

Example — identity swap:

Swap the face to match the reference photo. Keep the same expression and outfit.

The template merges this line into the GPT Image 2 prompt automatically. You do not need to write the long preservation block yourself.

6. Generate the restyled frame

Select 3. GPT Image 2 — Restyle Frame and click Generate.

Inspect results in the node sink—arrow through variants and pick the one where skin tone, hands, and text look correct. Regenerate if hands look older than the face or if text garbles.

Note: If you change the source video or extracted frame, re-run Extract First Frame, then GPT Image 2 again so downstream inputs stay in sync.

What you can change vs what stays locked

Safe to change (the look):

• Face / identity
• Clothing details and styling (when you ask explicitly)
• Background or environment (when you ask explicitly)
• On-screen text or logo (when you name the new text)

Usually stays the same (from the extracted frame):

• Pose and body position
• Framing and camera angle
• Overall lighting and cinematic grade
• Expression (copied from the source frame unless you ask otherwise)

The original video motion is preserved because you are editing a still that shares the clip’s timing anchor—not regenerating the whole video in one prompt.

Tips and troubleshooting

• Blurry or dark frames → scrub to another moment or use a brighter source clip.
• Hands look wrong → regenerate; the template already asks for matching skin age and tone across face and hands—sometimes it takes two passes.
• Text did not update → make the Additional edits line explicit: Change text to YOUR BRAND HERE.
• Nothing updates after you swap media → run nodes in order: Extract Frame → GPT Image 2.
• Ready for motion → take the approved still into Motion Apply with your source video as the motion driver.

Limitations

• This template stops at the restyled still. Full motion transfer is a separate step on Motion Apply.
• Extreme occlusions (heavy masks, turned-away faces) reduce swap quality—pick another frame first.
• Legal and brand review still apply for client work; generated text may need a human compliance check.

Next step

Open The Kumar Method in Explore, swap in your video and reference, and iterate until the still is approval-ready. For the motion half of the pipeline, continue with Motion Apply or the full Kumar Method video tutorial (speaking clip + beat-synced tail montage in one graph).

For a related multi-scene technique, see AI clips where your main person stays the same.